perf: Optimize codegen by ~20%

[robertt]

As our Prisma schema has grown (now 3,695 lines, 117 models, and 1,809 fields), typegraphql-prisma generation has become a significant bottleneck, taking ~50 seconds in CI and production builds. This has a noticeable impact on iteration velocity.

This PR improves schema generation time by ~20% (std. dev. 3.96% across 100s of benchmarks) through:

• Parallel processing in cpu-aware batches
• Fine-grained caching for models, enums, inputs, and outputs
• Project-specific cache isolation
• Memory-stable batch execution

We’re patch-packageing this internally for now but would love to see it upstream. Let me know if you have any feedback or
questions!

As an aside, here are some more architectural improvements I think would speed it up much more significantly but I didn't have time to get to. Happy to discuss more if you're interested.

• Implement AST pooling system to cache and reuse common TypeScript AST nodes
• Add parallel file I/O and memory pooling for field generation
• Incremental compilation support for faster subsequent generations